Advances in Radiative Transfer Modeling in Support of Satellite Data Assimilation

Download as pdf or txt
Download as pdf or txt
You are on page 1of 9

NOVEMBER 2007 WENG 3799

Advances in Radiative Transfer Modeling in Support of Satellite Data Assimilation


FUZHONG WENG
NOAA/NESDIS/Office of Research and Applications, Camp Springs, Maryland

(Manuscript received 30 March 2006, in final form 23 January 2007)

ABSTRACT

Development of fast and accurate radiative transfer models for clear atmospheric conditions has enabled
direct assimilation of clear-sky radiances from satellites in numerical weather prediction models. In this
article, fast radiative transfer schemes and their components critical for satellite data assimilation are
summarized and discussed for their potential applications in operational global data assimilation systems.
The major impediments to the fast radiative transfer schemes are highlighted and a call is made for broader
community efforts to develop advanced radiative transfer components that can better handle the scattering
from atmospheric constituents (e.g., aerosols, clouds, and precipitation) and surface materials (e.g., snow,
sea ice, deserts).

1. Introduction 1960 to 1995, the forecasts of the 36-h mean sea level
pressure over North America went from having no skill
Satellite observations of the atmosphere, land, and at all (i.e., being no more useful than a forecast based
oceans are now a major component of the environmen- on climatology), to being skillful 72% of the time. This
tal observing system, since they provide critically im- dramatic improvement of forecast skill has been di-
portant information to better understand and forecast rectly attributed to the launch, beginning in 1995, of an
short-term as well as climatic changes in weather. improved series of meteorological satellite instruments,
Through data assimilation techniques, the satellite ob- called the Advanced Microwave Sounding Unit (AMSU),
servations as well as other sources of atmospheric and whose data are assimilated in the more sophisticated
oceanic data, sampled at different times, intervals, and NWP models, which are run using more powerful com-
locations can be combined into a unified and consistent puters (Uccellini et al. 2001).
description of the atmospheric state. Global objective Satellite measurements (radiances) are directly as-
analyses are produced from these diverse observations similated into the NWP model using a computationally
in combination with the a priori knowledge of the efficient, optimum procedure, which minimizes the dif-
evolving atmospheric state as given by numerical ference between the radiances and those computed
weather prediction (NWP) models. By far, the greatest based on the NWP output of the atmospheric state. In
volume of data ingested into these numerical models is the absence of cloud absorption and atmospheric scat-
from satellite instruments, whose data have contributed tering from precipitation an accurately parameterized
to a dramatic improvement of forecast accuracy over radiative transfer model was used to assimilate satellite
the last 20 years. measurements into global NWP models for clear atmo-
Since the first meteorological satellite was launched spheric conditions (e.g., Eyre 1989; Garand et al. 2001).
in April 1960, much progress has been made in the However, in order to utilize the full capabilities of
utilization of satellite data in NWP models. This AMSU and other advanced instruments, which in-
progress is a result of improved satellite instruments, cludes all weather conditions, an accurate radiative
increase in computer power, and improvements in nu- transfer model is needed that incorporates cloud ab-
merical models and data assimilation techniques. From sorption and scattering from hydrometeors. Such a for-
ward model has been developed to enable the compu-
tation of both the satellite radiances and their change
Corresponding author address: Dr. Fuzhong Weng, NOAA/ with atmospheric variables (i.e., Jacobian) over a broad
NESDIS, Room 712, 5200 Auth Road, Camp Springs, MD 20746. range of wavelengths from the microwave to the infra-
E-mail: [email protected] red (Weng et al. 2005).
DOI: 10.1175/2007JAS2112.1

Unauthenticated | Downloaded 08/01/23 04:47 AM UTC

JAS4062
3800 JOURNAL OF THE ATMOSPHERIC SCIENCES—SPECIAL SECTION VOLUME 64

Direct assimilation of satellite cloudy radiances into improvements in direct satellite radiance assimilation
NWP models is a difficult problem, requiring very ac- under all weather conditions. These components are
curate variational schemes. The problem has been pur- being developed through National Aeronautics and
sued by some major centers. At the European Centre Space Administration (NASA)/NOAA/Department of
for Medium-Range Weather Forecasts (ECMWF), Defense (DoD) Joint Center for Satellite Data Assimi-
Moreau et al. (2004) developed a one-dimensional lation (JCSDA) Program and include the forward ra-
variational (1DVAR) system to assimilate the radi- diative transfer model and its Jacobian, fast gaseous
ances at the microwave window channel frequencies absorption models, and surface emissivity and reflectiv-
that are more sensitive to cloud liquid water, water ity models.
vapor, sea surface wind, and temperature. The 1DVAR
retrievals of the vertically integrated cloud liquid water 2. Current radiative transfer schemes used for
and water vapor are then assimilated into a four- satellite data assimilation
dimensional variation (4DVAR) system. A similar pro-
a. Discretization of radiative transfer model
cess has been also developed for 1DVAR SSM/I re-
trieval at Meteorological Service of Canada (MSC; De- Data assimilation of satellite data uses a variational
blonde et al. 2007). At National Oceanic and analysis approach to combine the satellite measure-
Atmospheric Administration (NOAA), Weng et al. ments with an a priori initial guess based on the NWP
(2007) also developed a Hybrid Variational Scheme model (Rodgers 2000). Specifically, assuming that the
(HVAR), which is a 1DVAR retrieval of temperature errors in the satellite observations, I0, and a priori in-
profiles from the AMSU instrument, and 4DVAR as- formation, I, are unbiased, uncorrelated, and have
similation of retrieved temperature profiles. A similar Gaussian distributions, the best estimate of the atmo-
algorithm has also been tested in the UK Met Office spheric state, x, minimizes the cost function:
(English and Une 2006). During the 2005 hurricane sea- 1
son, the HVAR was applied to several storms, result- J ⫽ 共x ⫺ xb兲TB⫺1共x ⫺ xb兲
2
ing in improved analyses of the hurricanes three-
dimensional warm core temperature structure and ac- 1
⫹ 关I共x兲 ⫺ Io兴T共E ⫹ F兲⫺1关I共x兲 ⫺ I0兴, 共1兲
companying wind fields within and around the rain 2
bands. Both the lower-level wind speed and upper-level where the second term contains the coupling of satellite
divergence were also enhanced, displaying a reasonable radiances with the state variables derived from NWP
asymmetric structure. models. In addition to the observed radiance vector, I0,
Until now much of the impact of satellite data has and simulated radiance vector, I, for a set of channels
been demonstrated through radiance assimilation of (or frequencies), (1) also contains the error covariance
cloud-free atmospheres. In the next decade, many ad- matrix, B, associated with the background state vari-
vanced microwave and infrared sensors will be de- able xb, and the error matrices associated with obser-
ployed in space with higher spatial and spectral resolu- vations, E and forward models, F.
tion so as to increase their sensitivity to aerosols, Minimization of the cost function is obtained using
clouds, precipitation, and surface parameters beyond an iterative process that computes the descent direction
that of current instruments. To utilize the data from at state x. The value of the cost-function gradient at
current sensors as well as the next generation of instru- each iteration is given by (Eyre 1989; Garand et al.
ments, the forward-modeling capability needs to be en- 2001)
hanced to include the scattering and polarization result-
ing from these atmospheric and surface features. Only ⵜxJ ⫽ B⫺1共x ⫺ xb兲 ⫹ HT共E ⴙ F兲⫺1关I共x兲 ⫺ I0兴, 共2兲
then will the assimilation of aerosol, clouds, and rain- where HT is the adjoint operator of the Jacobian matrix,
affected radiances make a major impact on NWP fore- H, which is the derivative of the radiance with respect to
casting, and add to our knowledge of clouds, air quality, the input variables (e.g., ⳵I/⳵x) as determined using a
and the hydrological cycle. Of course, to facilitate im- tangent-linear approximation of the radiative transfer
proved forecasting through direct radiance assimila- model. The radiative transfer model also provides the
tion, concomitant improvements must also be made in relationship between the model state vector and ob-
the microphysical modeling of hydrometeors and the served radiances [i.e., I(x)], by determining the radi-
use of higher NWP grid resolutions to resolve convec- ance output to state vector inputs.
tive-scale processes. For a plane-parallel atmosphere, the radiance vector
This paper will review several radiative transfer mod- is obtained from the differential form of the radiative
eling components that are critically important to the transfer equation

Unauthenticated | Downloaded 08/01/23 04:47 AM UTC


NOVEMBER 2007 WENG 3801


dI共␶, ␮, ␾兲
d␶
⫽ ⫺I共␶, ␮, ␾兲 ⫹

4␲ 冕 冕
2␲

0
1

⫺1
M共␶ ; ␮, ␾; ␮⬘, ␾⬘兲I共␶, ␮⬘, ␾⬘兲 d␮⬘ d␾⬘ ⫹ S共␶, ␮, ␾; ␮0, ␾0兲, 共3兲

where the integral term contains the contributions due to multiple scattering while the source term is given by

M11共␮, ␾; ␮0, ␾0兲

冤冥 冤 冥
1
0 ␼F0 M12共␮, ␾; ␮0, ␾0兲
S ⫽ 共1 ⫺ ␼兲B关T共␶兲兴 ⫹ exp共⫺␶Ⲑ␮0兲 , 共4兲
0 4␲ M13共␮, ␾; ␮0, ␾0兲
0 M14共␮, ␾; ␮0, ␾0兲

where M is the scattering phase matrix; I ⫽ [I, Q, U, For scattering media, numerical radiative transfer so-
V ]T is the Stokes vector; B(T ) is the Planck function at lutions require that the integral term in (3) be ex-
temperature T; F0 is the solar spectral constant; ␮0 pressed as a finite series so that the radiance field is
and ␾0 are the cosines of the zenith angle and the azi- represented as a number of discrete streams from dif-
muthal angle of the sun, respectively; ␮ and ␾ are the ferent directions, that is, zenith angles. Depending on
cosines of zenith angle and the azimuthal angle in the the observation angle and accuracy requirement, the
scattering direction, respectively; ␼ is the single- number of streams can be as little as two (Bauer et al.
scattering albedo; and ␶ is the optical thickness. The 2005). Furthermore, the azimuthal dependence of the
second term in (4) is generally omitted at microwave Stokes vector is expanded in a series of Fourier har-
and infrared frequencies but is occasionally used for monics whose amplitudes are a function of zenith angle
applications such as sun glint effects on the microwave (or streams). The amplitude for each zenith angle (or
measurements. stream) can then be simplified as

冋Im共␶, ␮i兲
册 冋
Im共␶, ␮i兲
册 兺冋
Mm共␮i, ␮j兲 Mm共␮i, ␮⫺j兲
册冋 Im共␶, ␮j兲
册 冋 Sm共␶, ␮i, ␮0兲

N
d
␮i ⫽ ⫺␼ wj ⫺ ,
d␶ ⫺Im共␶, ␮⫺i兲 Im共␶, ␮⫺i兲 j⫽1 Mm共␮⫺i, ␮j兲 Mm共␮⫺i, ␮⫺j兲 Im共␶, ␮⫺j兲 Sm共␶, ␮⫺i, ␮0兲

共5兲

where ␮i and wi are the Gaussian quadrature points and tween the radiance (i.e., brightness temperature) for
weights, respectively. Note that ␮⫺I ⫽ ⫺␮i and w⫺I ⫽ the two-stream and multistream model is typically less
wi. Because of the characteristics of the phase matrix, than 2 K at various microwave frequencies so that con-
the radiance components for the sinusoidal and cosinu- cise analytic solutions can be obtained (Liu and Weng,
soidal azimuthal modes can be decoupled, recombined, 2002). However, a more general class of scatterers must
and solved independently (Weng 1992; Schulz et al. be considered to characterize the polarization from
1999; Weng and Liu 2003). clouds at infrared and microwave wavelengths. To ob-
Equation (5) can be solved using standard routines tain solution for this case, a delta-four stream scheme
such as the multilayer discrete ordinate method (Weng was developed that reduces the phase function to four
1992; Schulz et al. 1999), the Doubling-adding method expansion terms (Liou et al. 2005).
(Evans and Stephens 1991), and the matrix operator Other fast schemes used to solve the radiative trans-
method (Liu and Ruprecht 1996). For spherical and fer Eq. (3) include the successive order of interaction
randomly oriented nonspherical scatters the scattered (Greenwald et al. 2005). This method takes advantage
radiation is azimuthal independent so that the integra- of the truncated doubling method for a single layer and
tion of the phase matrix over azimuth angle can be the successive order of interaction for vertical integra-
determined analytically (Liu and Weng 2002). Also, the tion. The truncation used in the doubling procedure is
first two Stokes components, I and Q, are decoupled a good approximation for infrared and microwave fre-
from the U and V components, and the solutions can be quencies, where errors in the brightness temperature
expressed in terms of the atmospheric and surface op- are generally less than 0.1 K when using 16 streams.
tical parameters for fast computation. Furthermore, for This error is significantly less than that derived with the
these particular scattering particles, the difference be- delta-Eddington model (Bauer et al. 2005) while the

Unauthenticated | Downloaded 08/01/23 04:47 AM UTC


3802 JOURNAL OF THE ATMOSPHERIC SCIENCES—SPECIAL SECTION VOLUME 64

computational speeds between the two models are very aerosol particles). The radiative transfer models now
similar. Another approach is the discrete ordinate tan- start including a treatment of completely or partially
gent-linear radiative transfer solution (Voronovich and cloud-filled fields of view and can be applicable for (a)
Gasiewski 2004), which uses symmetric matrices to aerosol layers both above or below cloud layers, (b)
make the solution numerically stable and simpler to multiple cloud layers (e.g., an ice- cloud layer over a
calculate. Lastly, a doubling-adding method (DA) is water-cloud layer) in a vertical column, and (c) a single-
one of the most accurate tools for obtaining detailed layered cloud containing a mixture of both ice- and
multiple-scattering calculations and has served as a ref- water-phase particles, or the so-called mixed-phase
erence for comparison of the other radiative transfer clouds.
solutions. However, the method has never been used For an ice-phase cloud, a comprehensive database of
operationally because of the tremendous demand on the single-scattering properties of ice particles with
computational resources. Liu and Weng (2006) derived various geometries has been developed (Yang et al.
an analytical expression to replace the more compli- 2005). For a liquid-phase cloud, the droplets are simu-
cated thermal source terms in the DA method, which is lated using a modified gamma size distribution with an
now referred as the advanced doubling-adding (ADA) effective radius as a function of rain rate for precipitat-
method. The ADA is about 61 times faster and has the ing clouds. For nonprecipitating clouds, cloud optical
same accuracy as the DA method. Computation codes properties are computed from cloud water content and
of ADA have been translated into the tangent-linear effective radius (Liu and Weng 2006). The dielectric
and adjoint codes for determining the linear operators constant of water is computed from a microwave model
contained in the cost function used for satellite data (Ulaby et al. 1986) and an infrared and visible model
assimilation [see (2)]. (Irvine and Pollack 1968. To quantify the effect of aero-
Each of the above models is valid at different wave- sols in the data assimilation system, a radiative transfer
lengths, and has different accuracies and computational model including aerosol is requested. In the current
speed. Intercomparisons such as that done by Smith et NOAA air quality forecast and satellite data assimila-
al. (2002) provide vital information on the optimal se- tion system, aerosol distributions and types are taken
lection of the radiative transfer and Jacobian model to from Goddard Chemistry Aerosol Radiation and
be used in satellite data assimilation. The advantages of Transport (GOCART) model that simulates major tro-
each model may ultimately be combined in a more re- pospheric aerosol components of dust, sulfate, black
fined version by taking advantage of their speed and carbon (BC), organic carbon (OC), and sea-salt aero-
accuracy relative to benchmark solutions. sols. The organic carbon and black carbon are further
One of critical issues affecting the speed of radiative distinguished with hydrophobic and hydrophilic. Under
transfer calculations for precipitating atmospheres is a hydrophilic situation, the aerosol effective size swells
the determination of the scattering coefficients and with ambient water vapor. The sea-salt aerosol is also
phase-matrix parameters for different types of hydro- divided into a small mode (SSAM) and a coarse mode
meteor distributions. Operational NWP assimilation (SSCM). The effective aerosol size for SSAM, SSCM,
permit only a fraction of a millisecond time for com- and sulfate depends on ambient relative humidity. The
plete calculation of the radiance and its Jacobian. How- size of dust aerosol is independent from relative humid-
ever, the term-by-term Mie calculations of the scatter- ity and its range is divided into 5 size bins in GOCART
ing coefficients and phase-matrix parameters are not model where the first size bin corresponds to 4 sub size
only too computationally intensive for operational use bins with fixed volume concentrations in percentage.
but often involve redundant calculations for similar hy- There are 8 aerosol types in total: dust, dry OC, wet
drometeor profiles. One way to overcome this problem OC, dry BC, wet BC, SSAM, SSCM, and sulfate. The
is to produce lookup tables for the parameters that output of the GOCART model is the concentrations
cover the range of hydrometeor types, size distribu- for up to 5 aerosol types of the 8 aerosol types. A
tions, and temperature conditions, and also fit within a lookup table is necessary to store the precalculated
small prescribed memory space. aerosol optical parameters such as dry mass extinction
To facilitate fast forward radiative transfer simula- coefficient, single-scattering albedo, and asymmetry
tions required for the data assimilation under com- factor. In the aerosol lookup table, a lognormal size
pletely or partially cloudy conditions, it is critical to distribution is characterized through a median radius
build advanced radiative transfer models that can ac- and its geometric standard deviation parameters (Chin
count efficiently and accurately for both the absorption et al. 2002). The refractive indices of these chemical
of atmospheric gases and the multiple scattering of at- components are based on studies by Hess et al. (1998).
mospheric particles (cloud droplets, ice crystals, and Lorenz–Mie code is used to compute mass extinction

Unauthenticated | Downloaded 08/01/23 04:47 AM UTC


NOVEMBER 2007 WENG 3803


coefficient, single-scattering albedo, asymmetry factor, N

and phase expansion coefficients. Ī ⫽


⌬␯
␾共␯兲I共␯兲 d␯ ≅ 兺 w I共␯ 兲;
i⫽1
i i ␯i ∈ ⌬␯, 共6兲

b. Fast gas absorption models where N is the total number of nodes within a channel
spectrum domain with i being the ith node; ␾ is the
Under clear atmospheric conditions, radiative trans-
spectral response function; ␯i is the wavenumber node
fer modeling uses atmospheric absorption coefficients and wi is the weight determined by fitting the “exact”
as the key input. The absorption varies with the atmo- calculations (from LBL model) for globally represen-
spheric conditions in a complicated way and is often tative set of atmospheres (training set). At the selected
computed through the line-by-line (LBL) spectral mod- nodes, the lookup tables of absorption coefficients for
els. Although LBL models are accurate, they take con- relevant species are stored. Maximum brightness tem-
siderable time to calculate transmittances for just a few perature errors between the LBL and OSS using the
atmospheres. To provide accurate transmittances in a lookup table calculations are less than 0.05 K in the
timely fashion, fast optical transmittance (OPTRAN) infrared and are about 0.01 K at microwave wave-
models have been developed for specific instrument lengths (Moncet et al. 2004).
channels. The current fast models used in the U.S. op- In the NWP community, there are many fast gaseous
erational data assimilation are based on an approach absorption models developed for satellite data assimi-
developed by McMillin and Fleming (1976), but there lation. A complete survey and the performance accu-
have been incremental changes made by a number of racy of these models are reviewed by Garand et al.
groups since then. With the coming of hyperspectral (2001). It is shown that for specific instruments like the
instruments, the requirements placed on the fast mod- Atmospheric Infrared Spectrometer (AIRS), a Stand
els are changing, and this modification process is con- Alone Radiative Transfer model (SARTA) developed
tinuing. Radiative transfer process should take into ac- by University Maryland at Baltimore County (UMBC;
count the instrument filter response functions because Strow et al. 2003) performs the best. This was obtained
each new instrument flown on a new spacecraft has its by comparing SARTA with all other fast models that
own specifications. Appropriate changes are made to were not specifically fitted to the AIRS instrument. Un-
the LBL models to regenerate the new coefficients for fortunately, the original version of SARTA does not
the transmittances of new instruments. have tangent-linear and adjoint models.
Recently, a fast and optimal spectral sampling (OSS) The transmittance at microwave frequencies near 60
absorption model was developed for improved accu- GHz is also highly affected by the Zeeman effects,
racy across the entire spectrum (Moncet et al. 2004). which split the oxygen spectral lines near 60 GHz into
The OSS is a new approach that allows for rapid cal- a number of finely spaced lines. The intensity of the
culations of radiance for any class of multispectral, hy- split lines, and their separation, depends on the earth’s
perspectral, or ultraspectral sensors at any spectral magnetic field strength and its orientation with respect
resolution and any wavelength by selecting and appro- to the satellite instrument’s viewing direction and its
priately weighting the monochromatic radiances con- received polarization. For fast computation of the ab-
tributed from gaseous absorption and particle scatter- sorption coefficients including the Zeeman effects, a
ing over the sensor bandwidth. This allows the calcula- LBL model (Rosenkranz 1995) is parameterized using
tion to be performed at a small number of spectral various predictors including the earth magnetic field
points while retaining the advantages of a monochro- magnitude (B), polarization (left and right circularly),
matic calculation such as that needed for the exact temperature, and angle between magnetic field and
treatment of multiple scattering and/or polarization. propagation direction of electromagnetic (EM) wave.
The OSS method is well suited for remote sensing ap- This parameterization has resulted in an excellent simu-
plications that require extremely fast and accurate ra- lation of the brightness temperatures from the Defense
diative transfer calculations: atmospheric compensa- Meteorology Satellite Program (DMSP) Special Sensor
tion, spectral and spatial feature extraction, multisensor Microwave Imager and Sounder (SSMIS; Han 2006).
data fusion, subpixel spectral analysis, qualitative and
quantitative spectral analysis, sensor design, and data
c. Surface emissivity modeling
assimilation. The OSS is currently used as part of the
National Polar-Orbiting Operational Environmental For window channels as well as lower sounding chan-
Satellite System (NPOESS) environmental parameter nels, the measurements respond to the radiation ema-
retrieval algorithms. With the OSS method, the channel nating from the earth’s surface. As such, the radiance
radiance is calculated from and Jacobian computations require accurate knowledge

Unauthenticated | Downloaded 08/01/23 04:47 AM UTC


3804 JOURNAL OF THE ATMOSPHERIC SCIENCES—SPECIAL SECTION VOLUME 64

of the surface emissivity and reflectivity. Without accu- ness is approximated using small perturbation theory.
rate surface models, the measurements from these However, it soon became evident that these emissivity
channels are often greatly different than the a priori models were deficient in representing the true surface
estimate, so that the radiances cannot be assimilated features retrieved from the AMSU instrument (Weng
into NWP models. Therefore, in addition to the radia- et al. 2001) as well as the aircraft microwave measure-
tive transfer model of the atmosphere, a surface model ments obtained during the NASA Cold Land Processes
must be developed to properly include the variability of (CLP) experiments (Stankov et al. 2004). Furthermore,
emissivity and reflectivity. in cold climate regimes such as Greenland, the internal
Shortly after the launch of the first AMSU instru- characteristics of snow are very complex, exhibiting
ment in 1998, an ocean emissivity model was developed both stratification and metamorphosis. As such, very
at NOAA/National Environmental Satellite, Data, and large differences exist between the satellite measure-
Information Service (NESDIS) for radiance assimila- ments and modeled emissivity.
tion. This model calculated the emissivity by linear Modeling the land emissivity at infrared wavelengths
combining of the contributions from calm and rough is even less advanced, compared to that at the micro-
water. The calm water emissivity was calculated using wave frequencies. Much of this stems from the fact that,
the Fresnel equations with the dielectric constant for compared to microwave frequencies, the infrared emis-
seawater as input (Klein and Swift 1977), while the sivity is generally higher and less variable over most
emissivity for wind-driven seas was calculated using the land surfaces. However, over deserts, minerals such as
empirical model initially developed by Stogryn (1972), quartz have emissivities near unity while the emissivity
which includes the foam coverage and its relationship of limestone is near 0.6 for wavelengths between 7 and
to wind speed, viewing angle, and microwave frequency 9 ␮m. Instead of relying on emissivity models, a lookup
(Hollinger 1971; Wilheit 1979). A sea surface emissivity table of IR emissivity derived from satellite and
model developed by the Met Office, United Kingdom ground-based retrievals (Knuteson et al. 2006) is pro-
(English and Takashima 1998), was also tested in the posed for uses in satellite data assimilation systems. In
operational environment. This model was initially de- the current NOAA global operational data assimilation
veloped based on geometrical optics, where the large- system, the emissivity spectra are specified as function
scale ocean waves are modeled using tilting surface fac- of surface types.
ets whose scattering coefficients are proportional to the
number of surface facets with a sloping angle satisfying
3. Major impediments
the specular reflection condition. The slope distribution
of the large-scale roughness was computed from an Direct assimilation of satellite radiances under clouds
ocean surface spectrum (Cox and Munk 1954). It is and precipitation requires detailed information on the
found that the first version of the UK fast emissivity profiles of cloud microphysical variables as background
model (FASTEM-1) produced better results in simu- information and the error characteristics of the error
lating the AMSU data at higher frequencies, whereas covariance matrix as shown in (1) and (2). Currently,
the NEDSIS emissivity model performs better at fre- operational forecast models and cloud prognostic
quencies less than 37 GHz. schemes run slightly different physical packages. In
Microwave emissivity models were also developed to principle, this prediction scheme can resolve cloud con-
help improve the radiance assimilation over land densates only when the model resolution is increased to
(Weng et al. 2001). Prior to this model development, less than a few kilometers. At courser resolutions, the
constant emissivity values were used for unfrozen land, forecast model has to rely on a cumulus parameteriza-
snow cover, and sea ice in the NOAA global data as- tion scheme (Arakawa and Schubert 1974; Moorthi and
similation system. Modeling the emissivity for such het- Suarez 1999) to determine the clouds and precipitation
erogeneous surfaces is a daunting task. In the case of associated with convective motion. In addition to the
snow it requires an understanding of radiative transfer cumulus parameterization, NWP models also employ a
theory for dense media (Weng et al. 2001). For ex- stratiform cloud water parameterization scheme (e.g.,
ample, a more physically based emissivity model was Sundqvist 1978).
developed for snow, which includes the volumetric scat- To understand the quality of the model-predicted
tering from ice crystals based on strong fluctuation cloud condensates, more observational datasets to
theory. In the case of vegetation-covered land, geo- characterize the errors of the forecasted cloud water–
metrical optics was used to calculate the leaf reflectivity ice content need to be developed. Retrievals from sat-
and transmissivity since the leaf size is typically larger ellite passive sensors may be used for assessments of
than the microwave wavelength while surface rough- model errors in the column-integrated water (Weng et

Unauthenticated | Downloaded 08/01/23 04:47 AM UTC


NOVEMBER 2007 WENG 3805

al. 1997). However, errors in the profiles of cloud con- probably extracted during the data assimilation pro-
densates predicted by forecasting models may be better cess. Most recently, various fast gas absorption models
characterized when the data from aircraft in situ mea- have been quantified for their errors relative to the
surements or satellite active sensors such as CloudSat LBL computations (Saunders et al. 2007). A compari-
(Stephens et al. 2002) are available. Also, the satellite- son of radiative transfer models for simulating radi-
derived cloud water is subject to substantial retrieval ances with AIRS has been also undertaken. Results
errors. from 14 LBL and fast parameterized infrared models
Further difficulties and limitations arise from the fact agree to within 0.02 K for forward computations when
that scattering by clouds and precipitation is a function compared to a reference LBL model averaged over a
of particle size, which is not currently predicted in NWP subset of profiles and regions, and the mean differences
models. This problem is acute when the size parameter increase to 0.2 K when compared with AIRS observa-
(ratio of particle size to wavelength) is large. Determi- tions. For the Jacobians, the gradients of radiance to
nation of mean particle size from satellite visible/near the state parameters from all models have some pro-
infrared (VIS/NIR) or development of diagnostic and/ files/channels that do not fit the reference well (Saun-
or prognostic schemes in NWP models for mean par- ders et al. 2007).
ticle size is crucial for the use of cloudy radiances in The performance of forward models under scattering
NWP models. Alternatively, an ad hoc relationship be- cloudy and precipitating atmospheres can vary substan-
tween cloud particle size and ambient temperature and tially from case to case. At visible wavelengths, errors
ice water content (Heymsfield and Platt 1984) can be are strongly dependent on our knowledge on particle
also tested for radiative transfer in ice-phase clouds. shapes and orientation, and spatial inhomogeneity. In
At the infrared wavelength ranging from 3 to 15 ␮m, infrared and microwave wavelengths, the errors can
aerosols may have considerable effect on radiance cal- range from a few tenth to tens of kelvins. In aerosol-
culations. It is known that most models developed for scattering conditions, for example, simulated High
chemistry aerosol radiation and transport take into ac- Resolution Infrared Sounder (HIRS) radiances can be
count the major tropospheric aerosol components, in- biased by 2–3 K at its window channels (Liu et al. 2007).
cluding sulfate, dust, black carbon, organic carbon, and Over land where surface emissivity varies, the errors
sea-salt aerosols. Each of these aerosols requires more are strongly dependent on our knowledge in surface
demanding computational power for determining their emissivity and atmospheric conditions. Taking micro-
optical parameters. In addition, knowledge of their con- wave water vapor sounding channels as an example, the
centrations, particle size, and shape is very limited from contributions to the brightness temperature from sur-
any ground-based measurements, although they are face emission and reflected downwelling radiation can
critical parameters in radiative transfer calculations. be significantly large over polar regions where atmo-
For fast gas absorption models, more trace gases such spheres are typically dry and become semitransparent.
as carbon dioxide, monoxide and methane should also In the absence of scattering, an error of emissivity of a
be considered, because forecast models are making few percent can result in an error in brightness tem-
more use of satellite measurements that are sensitive to peratures of a few kelvins at 183.3 ⫾ 3 and 183.3 ⫾ 7
these constituents. Presently, transmittance models GHz, respectively, (note that these channels are nomi-
only include a number of “fixed” gases and variable nal sounding channels for most of atmospheric condi-
gases such as water vapor and ozone. Assimilating the tions; Weng and Yan 2003).
satellite measurements into forecasting models and pre-
dicting their distributions require the transmittance
models that include variations in minor gases. As the 4. Future work
gas absorption models and other radiative transfer
For the radiative transfer component, the highest pri-
components become more accurate, the variations in
ority items are as follows:
retrieved temperature due to changes in minor gases
become significant when they are ignored. 1) Continue refining the rapid gas absorption model for
Quantitative assessments of errors arising from vari- current and future sensors, including the absorption
ous radiative transfer modules are very important com- coefficients for hyperspectral instruments such as
ponents because the errors are typically used to define AIRS, Infrared Atmospheric Sounding Interferom-
the observational error covariance matrices in satellite eter (IASI), Cross-track Infrared Sounder (CrIS),
data assimilation systems. Without knowledge of errors and the corresponding Jacobian codes. Developed
of simulated radiances relative to observations, the in- models should be in line with the current data as-
formation from new satellite measurements cannot be similation interface at the Joint Center for Satellite

Unauthenticated | Downloaded 08/01/23 04:47 AM UTC


3806 JOURNAL OF THE ATMOSPHERIC SCIENCES—SPECIAL SECTION VOLUME 64

Data Assimilation (JCSDA) community radiative of the sea surface from photographs of the sun’s glitter. J.
transfer model. Opt. Soc. Amer., 44, 838–850.
Deblonde, G., J.-F. Mahfouf, B. Bilodeau, and D. Amselmo, 2007:
2) Update rapid atmospheric gaseous absorption coef-
One-dimensional variational assimilation of SSM/I observa-
ficients as instrumental parameters, spectral knowl- tions in rainy atmospheres at MSC. Mon. Wea. Rev., 135,
edge, or requirements change. For infrared sound- 152–172.
ing instruments that experience changes in the spec- English, S. J., and T. J. Takashima, 1998: A fast generic millime-
tral response shape and central wavelength, fast ter-wave emissivity model. Proc. SPIE, 3503, 288–300.
models are needed to rerun the LBL forward-model ——, and U. Une, 2006: Assimilation of AMSU cloudy radiances.
Preprints, 15th Int. TOVS Study Conf., Maratea, Italy, Int.
calculation with the improved instrumental response
TOVS Working Group, 447–454.
functions and new fast model coefficients. Evans, K. F., and G. L. Stephens, 1991: A new polarized atmo-
3) Develop and improve radiative transfer schemes to spheric radiative transfer model. J. Quant. Spectrosc. Radiat.
include scattering from aerosol, clouds, and precipi- Transfer, 46, 413–423.
tation. One of the key components is to speed up the Eyre, J. R., 1989: Inversion of cloudy satellite sounding radiances
computation with a lookup table of optical param- by non-linear optimal estimation. I: Theory and simulation
for TOVS. Quart. J. Roy. Meteor. Soc., 115, 1001–1026.
eters using particle size, dielectric constant, and mix-
Garand, L., and Coauthors, 2001: Radiance and Jacobian inter-
ing ratio as inputs. comparison of radiative transfer models applied to HIRS and
4) Develop infrared emissivity models for land sur- AMSU channels. J. Geophys. Res., 106, 24 017–24 032.
faces. Surfaces containing quartz, limestone, and Greenwald, T., R. Bennartz, C. O’Dell, and A. Heidinger, 2005:
other mineral compositions result in large variability Fast computation of microwave radiances for data assimila-
in the infrared emissivity at wavelengths from 3 to 4 tion using the “successive order of scattering” method. J.
Appl. Meteor., 44, 960–966.
and 6 to 9 ␮m, and should be treated using mixing
Han, Y., 2006: Fast RT models for SSMIS upper air sounding
formulas having variable composition. channels. Preprints, 15th Int. TOVS Study Conf., Maratea,
5) Improved microwave dense media theory is needed Italy, Int. TOVS Working Group, 25.
to simulate the emissivity for snow and sea ice. Hess, M., P. Koepke, and I. Schult, 1998: Optical properties of
Dense media scattering should also be developed to aerosols and clouds: The software package OPAC. Bull.
Amer. Meteor. Soc., 79, 831–844.
include vertical stratification and solved through a
Heymsfield, A. J., and C. M. R. Platt, 1984: A parameterization of
comprehensive algorithm that is applicable to a the particle size spectrum of ice clouds in terms of ambient
broad range of wavelengths. The inputs to the snow temperature and ice water content. J. Atmos. Sci., 41, 846–
and sea ice emissivity models should be closely 855.
linked to the output parameters from NWP bound- Hollinger, J. P., 1971: Passive microwave measurements of sea
ary layer models. surface roughness. IEEE Trans. Geosci. Electron., GE-9,
165–169.
Irvine, W. M., and J. B. Pollack, 1968: Infrared optical properties
Acknowledgments. The author would like to express
of water and ice sphere. Icarus, 8, 324–360.
sincere thanks to Professors Kuo-Nan Liou (UCLA) Klein, L. A., and C. T. Swift, 1977: Emissivity for calm water.
and Albin Gasiewski (University of Colorado) for their IEEE Trans. Antennas Propag., 25, 104–111.
comments on the manuscript. The views and opinions Knuteson, R., H. Rovercomb, and T. Tobin, 2006: Land surface
contained in this paper reflect those of the author and temperature derived from the advanced IR sounders on
should not be construed as an official National Oceanic METOP and NPOESS. Extended Abstracts, First ITWG Re-
mote Sensing and Modeling of Surface Properties, Paris,
and Atmospheric Administration or U.S. Government France, Int. TOVS Working Group.
position, policy, or decision. Liou, K.-N., S. C. Ou, Y. Takano, and Q. Liu, 2005: A polarized
delta-four-stream approximation for infrared and microwave
REFERENCES radiative transfer: Part I. J. Atmos. Sci., 62, 2542–2554.
Liu, Q., and E. Ruprecht, 1996: A radiative transfer model: Matrix
Arakawa, A., and W. H. Schubert, 1974: Interaction of cumulus operator method. Appl. Opt., 35, 4229–4237.
cloud ensemble with the large scale environment. Part I. J. ——, and F. Weng, 2002: A microwave polarimetric two-stream
Atmos. Sci., 31, 674–701. radiative transfer model. J. Atmos. Sci., 59, 2396–2402.
Bauer, P., E. Moreau, F. Chevallier, and E. O’Keffee, 2005: Mul- ——, and ——, 2006: Advanced doubling–adding method for ra-
tiple scattering microwave radiative transfer for data assimi- diative transfer in planetary atmosphere. J. Atmos. Sci., 63,
lation applications. Quart. J. Roy. Meteor. Soc., 131, 1259– 3459–3465.
1281. ——, Y. Han, P. van Delst, and F. Weng, 2007: Modeling aerosol
Chin, M., and Coauthors, 2002: Tropospherical aerosol optical radiance for NCEP data assimilation. Fourier Transform
thickness from the GOCART model and comparisons with Spectroscopy and Hyperspectral Imaging and Sounding of the
satellite and sun photometer measurements. J. Atmos. Sci., Environment, Optical Society of America, CD-ROM.
59, 461–483. McMillin, L., and H. Fleming, 1976: Atmospheric transmittance
Cox, C. S., and W. H. Munk, 1954: Measurement of the roughness model of an absorbing gas: A computationally fast and accu-

Unauthenticated | Downloaded 08/01/23 04:47 AM UTC


NOVEMBER 2007 WENG 3807

rate model for absorbing gases with constant mixing ratios in Sundqvist, H., 1978: A parameterization scheme for non-
inhomogeneous atmospheres. Appl. Opt., 15, 358–363. convective condensation including prediction of cloud water
Moncet, J., G. Uymin, and H. E. Snell, 2004: Atmospheric radi- content. Quart. J. Roy. Meteor. Soc., 104, 677–690.
ance modeling using the Optimal Spectral Sampling (OSS) Uccellini, L., F. Einaudi, J. Purdom, D. Rogers, R. Gelaro, J.
method. Preprints, SPIE Defense and Security Symp., Conf. Dodge, R. Atlas, and S. Lord, 2001: Weather prediction im-
5425: Algorithms and Technologies for Multispectral, Hyper- provement using advanced satellite technology. Preprints,
spectral, and Ultraspectral Imagery X, Orlando, FL, Society of 11th Conf. on Satellite Meteorology and Oceanography, Madi-
Photo-Optical Instrumentation Engineers, 5425–5437. son, WI, Amer. Meteor. Soc., CD-ROM, 3.1.
Moorthi, S., and M. J. Suarez, 1999: Documentation of version 2 Ulaby, F. T., R. K. Moore, and A. K. Fung, 1986: Microwave re-
of relaxed Arakawa–Schubert cumulus parameterization mote sensing, active and passive. From Theory to Applica-
with convective downdrafts. NOAA Tech. Rep. NWS/NCEP tion, Vol. III, Artech House, 2020–2027.
99-01, 44 pp. Voronovich, A. G., and A. J. Gasiewski, 2004: A fast multistream
Moreau, E., P. Lopez, P. Bauer, A. M. Tompkins, M. Janisková, scattering-based Jacobian for microwave radiance assimila-
and F. Chevallier, 2004: Variational retrieval of temperature tion. IEEE Trans. Geosci. Remote Sens., 42, 1749–1761.
and humidity profiles using rain rates versus microwave Weng, F., 1992: A multi-layer discrete-ordinate method for vector
brightness temperatures. Quart. J. Roy. Meteor. Soc., 130, radiative transfer in a vertically-inhomogeneous, emitting
827–852. and scattering atmosphere. I: Theory. J. Quant. Spectrosc.
Rodgers, C. D., 2000: Inverse Methods for Atmospheric Sounding: Radiat. Transfer, 47, 19–33.
Theory and Practice. World Scientific Publishing Company, ——, and Q. Liu, 2003: Satellite data assimilation in numerical
200 pp. weather prediction models. Part I: Forward radiative transfer
Rosenkranz, P. W., 1995: A rapid atmospheric transmittance al- and Jacobian models in cloudy atmospheres. J. Atmos. Sci.,
gorithm for microwave sounding channels. IEEE Trans. 60, 2633–2646.
Geosci. Remote Sens., 33, 1135–1140.
——, and B. Yan, 2003: A microwave snow emissivity model.
Saunders, R., and Coauthors, 2007: A comparison of radiative
Preprints, 13th Int. TOVS Study Conf., Quebec, QC, Canada,
transfer models for simulating Atmospheric Infrared
Int. TOVS Working Group, 212–219.
Sounder (AIRS) radiance. J. Geophys. Res., 112, D01S90,
——, N. Grody, R. Ferraro, Q. Zhao, and C. Chen, 1997: Global
doi:10.1029/2006JD007088.
cloud water distribution derived from Special Sensor Micro-
Schulz, F. M., K. Stamnes, and F. Weng, 1999: An improved and
wave Imager/Sounder and its comparison with GCM Simu-
generalized discrete ordinate radiative transfer model for po-
lation. Adv. Space Res., 19, 407–411.
larized (vector) radiative transfer computations. J. Quant.
Spectrosc. Radiat. Transfer, 61, 105–122. ——, B. Yan, and N. C. Grody, 2001: A microwave land emissivity
Smith, E., P. Bauer, M. F. Marzano, C. D. Kummerow, D. McK- model. J. Geophys. Res., 106, 20 115–20 123.
aque, A. Mugnai, and G. Panegrossi, 2002: Intercomparison ——, Y. Han, P. van Delst, Q. Liu, T. Kleespies, B. Yan, and L.
of microwave radiative transfer models for precipitating Marshal, 2005: JCSDA Community Radiative Transfer
clouds. IEEE Trans. Geosci. Remote Sens., 40, 541–549. Model. Preprints, 14th Int. TOVS Study Conf., Beijing,
Stankov, B. B., A. J. Gasiewski, B. L. Weber, M. Klein, R. Kelly, China, Int. TOVS Working Group, 217–222.
D. Cline, and G. A. Wick, 2003: Airborne measurement of ——, T. Zhu, and B. Yan, 2007: Satellite data assimilation in
snow cover properties using the Polarimetric Scanning Radi- numerical weather prediction models. Part II: Uses of rain-
ometer during the 2002 Cold Land Processes Experiment affected radiances from microwave observations for hurri-
(CLPX02). Proc. Int. Geoscience and Remote Sensing Symp., cane vortex analysis. J. Atmos. Sci., 64, 3910–3925.
Toulouse, France, 683–685. Wilheit, T. T., 1979: A model for the microwave emissivity of the
Stephens, G. L., and Coauthors, 2002: The CloudSat mission and ocean’s surface as a function of wind speed. IEEE Trans.
the A-Train. Bull. Amer. Meteor. Soc., 83, 1771–1790. Geosci. Electron., GE-17, 244–249.
Stogryn, A., 1972: A study of radiometric emission from a rough Yang, P., H. Wei, H.-L. Huang, B. A. Baum, Y. X. Hu, G. W.
sea surface. NASA Contractor Rep. NASA CR-2088, 80 pp. Kattawar, M. I. Mishchenko, and Q. Fu, 2005: Scattering and
Strow, L., S. Hannon, S. Machado, H. Motteler, and D. Tobin, absorption property database for nonspherical ice particles in
2003: An overview of the AIRS radiative transfer model. the near- through far-infrared spectral region. Appl. Opt., 44,
IEEE Trans. Geosci. Remote Sens., 41, 303–313. 5512–5523.

Unauthenticated | Downloaded 08/01/23 04:47 AM UTC

You might also like